A training method for SpikeProp without redundant spikes - Removing unnecessary sub-connections during training.
Takutoshi NakayamaTakashi MatsumotoHaruhiko TakaseHiroharu KawanakaShinji TsuruokaPublished in: ICNC-FSKD (2016)
Keyphrases
- training process
- training phase
- training algorithm
- feed forward neural networks
- detection method
- training samples
- high accuracy
- computational complexity
- synthetic data
- computationally efficient
- probabilistic model
- similarity measure
- experimental evaluation
- cost function
- training speed
- dynamic programming
- detection algorithm
- error rate
- pairwise
- high precision
- supervised training
- support vector machine
- preprocessing
- training set
- optimization method
- cross validation
- prior knowledge
- clustering method
- optimization algorithm
- input data
- denoising
- semi supervised